Position-based Content Attention for Time Series Forecasting with Sequence-to-sequence RNNs

TL;DR

This paper introduces an extended position-based attention mechanism for sequence-to-sequence RNNs, enhancing their ability to capture periodic patterns in time series and achieving state-of-the-art forecasting accuracy.

Contribution

It presents a novel attention model that can be integrated with any RNN to better model periodicities in time series data.

Findings

Achieves state-of-the-art forecasting performance on multiple datasets.

Effectively captures periodic and pseudo-periodic patterns.

Applicable to both univariate and multivariate time series.

Abstract

We propose here an extended attention model for sequence-to-sequence recurrent neural networks (RNNs) designed to capture (pseudo-)periods in time series. This extended attention model can be deployed on top of any RNN and is shown to yield state-of-the-art performance for time series forecasting on several univariate and multivariate time series.